Climate Shocks and Human Capital
The Impact of Natural Disasters on Students’ Performance in Standardized Tests
Mauricio Giovanni Valencia Amaya
1Abstract
Using a difference-in-difference approach with repeated cross-sections, this paper investigates the impact of the climate shocks occurred in Colombia in 2010 on the results in the country’s state standardized test, “Pruebas Saber 11”, during the period 2010-2012. Even though cognitive skills variables have been recognized as better proxies for human capital than quantitative measures, the literature on the relationship between climate shocks and human capital has focused on the latter. By using two unique datasets, one linking test scores with students’ socioeconomic characteristics, and the other containing the climate-related events at the municipal level, this paper contributes to the literature by providing a better estimate of the human capital costs of climate shocks. The findings indicate that the climate shocks of 2010 had a strong negative impact on the test outcomes, especially on those of low-income students; the impact was stronger on male and urban scholars; moreover, having experienced previous shocks seems to have lessened this impact, which points to the importance of adaptation and coping strategies; finally, health deterioration and physical capital destruction could have been two of the channels of transmission, due to the increase in the number of malaria and dengue cases, diseases that are related to weather conditions, and the damaged of school buildings, which might prevent students from attending classes.
Key words: Climate Shocks, Natural Disaster, Human Capital, Cognitive Skills, Colombia.
JEL codes: O12, I20.
Introduction
Using two unique datasets, and applying a difference-in-difference framework with repeated
cross-sections, this paper investigates the impact of the severe weather shocks that affected
Colombia in 2010-2011 on the results in the country’s state standardized test “Pruebas Saber 11” in
the period 2010-2012. Understanding the factors behind the performance of students in national
standardized tests is important since these results operate as a market signaling of the student’s
skills and knowledge; they also allow some students to continue studying at a higher education
level, as some universities not only require the test scores as part of their application process but
also use them to rank students, and so they are important for promoting social mobility.
1 Email: [email protected]. The author would like to thank Adriana Camacho (thesis
supervisor) Marcela Eslava, Catherine Rodríguez, Fabio Sánchez, and Alexis Munari for their valuable comments, and Guberney Muñetón for mapping some of the variables shown in this paper. The author is solely responsible for any errors or mistakes that may remain.
By using a qualitative proxy of human capital, such as cognitive test results, this paper will
contribute to the literature on the relationship between human capital and natural shocks. This
literature has focused primarily on quantitative proxies of human capital, such as years of schooling,
school enrollment ratios, students’ attendance or adult literacy rates. However, recent research on
human capital and educational systems is focusing more and more on qualitative measures of
educational attainment, such as cognitive skills (test score results), at the individual level, or a
country’s quality of education, at the aggregate level, rather than on quantitative measures. This is
because qualitative measures seem to be better predictors of economic growth and income
distribution, but also of individual’s future career success and productivity (Wößmann, 2003;
Orazem, 2007; Baird, 2012). Moreover, time spent in school does not necessarily translate into
more knowledge or better skills, since this variable is not a schooling outcome, but a component of
the educational production process (Orazem, 2007). In fact, differences in adult earnings are better
explained by cognitive achievements than by years of schooling (Glewwe, 2002, cited by Orazem
2007), as suggested by evidence for the United States and the United Kingdom (de Coulon et al.,
2011). Plus, cognitive tests results account for differences in the quality of education, one of the
cornerstones in the theory of human capital (Wößmann, 2003).
As argued by Orazem (2007), the use of measures of learning attainment in economics is still
a nascent field, subject to the availability of periodic academic datasets linking student’s scores with
student and family characteristics. So, previous studies on the relationship between climate shocks
and human capital have analyzed the impact of natural disasters only on quantitative indicators of
human capital; but so far there have been no studies to account for the effects of these disasters on
quality indicators of education. There is also a lack of studies on this relationship for Colombia,
although this country has suffered from several natural disasters in its past. In this sense, the use of
two unique datasets for Colombia, ICFES dataset and SNPAD dataset, allows this paper to measure
the impact of natural shocks on a qualitative proxy of human capital, such as learning attainment
(cognitive skills). ICFES dataset comprises “Saber 11” test scores (a national standardized test,
similar to SAT) plus the personal characteristics and family background of each individual
test-taker; and SNPAD dataset provides information on the natural disasters that have affected
Colombia’s municipalities since 1998.
The theory of human capital, proposed in the 1950s and 1960s by Schultz, Mincer, and
Becker, brought to the spotlight of development economics the importance of education and
accumulation of knowledge. From this theoretical point of view, education can be considered as an
education (micro level), but also to nations as a whole (macro level). In fact, the theory states that a
great part of the differences in wages are due to differences in the productivity of individuals, but
productivity is itself determined by previous investments made by these individuals in education
and training (Cahu and Zylberberg, 2004). Human capital is then a key element in the development
of nations; it enhances the welfare and the choices available to people (Nahapiet, 2011), but at the
same time it fosters economic growth (Barro, 2001). As Becker states:
The 21st century is clearly placing much greater emphasis than ever before on the importance of knowledge and information to the development of both countries and individuals (…) This means that it is more important than ever for both individuals and for nations to acquire knowledge, skills, and the experience to know how to acquire additional information (Becker, 2011, p. xv).
However, the stock and the accumulation of human capital can be threatened by the
uncertainty of climate shocks. These shocks, similar to economic downturns, will have an influence
not only on the returns of education for the people affected by the shock, but also on their attitudes
towards acquiring human capital (Broomhall and Johnson, 1994). The issue of natural disasters is
even more relevant to human capital if we consider the state of education in developing countries.
In these countries, the limited capacity in human and financial resources is one of the main reasons
why the quality of education is so low, with students learning much less than they should, according
to their curriculums, and also learning less compared to students in developed countries (Glewwe
and Kremer, 2006). In this scenario, climate shocks will make the convergence of these countries to
the quality standards in education reached by the developed world even more difficult.
Over the last hundred years the world has experienced a serious threat to its existing forms of
living: a significant warming, as a result of the increase in the emission of greenhouse gases. This
phenomenon has had regional and global consequences, such as: reduced soil moisture,
precipitation, droughts, sea level rise, high-temperature events, and floods, among others. Models
projections conclude that the average annual global temperature for the period between 2007-2027
will rise about 0.2°C per decade, the average sea level will increase by 0.1 to 0.2 meters by
2090-2099 (relative to 1980-1999), and the frequency of climate shocks, such as heavy precipitations, hot
extremes, and heat events, are very likely to increase in the next 100 years (IPCC, 2007). Moreover,
the wide range of those projections generates uncertainty in terms of the regional and local
socio-economic impacts of climate change (Yohe and Schlesinger, 2002).
Even though climate shocks are part of the history of mankind, the increasing rate of
occurrence of such events and its devastating effects on the lives of millions of people around the
percentage of the world population still depends on agriculture as its primary source of income, and
many of the people living in the surroundings of the urban areas (slum dwellers) live in deplorable
conditions. For these people, even small changes in climate can have an enormous impact, due to
the nonlinearity response of economic and social systems (Sachs, 2006). Climate shocks might
destroy crops and negatively affect not only people’s assets and savings, but also their health,
nutrition, and education. This implies uncertainty, vulnerability, and fewer opportunities to
overcome their current living conditions, creating cumulative vicious cycles of disadvantages that
are transmitted from generation to generation (UNDP, 2007).
What is more, those risks and vulnerabilities related to climate change are increasingly faced
by poor people (UNDP, 2007; The World Bank, 2010). Indeed, during the period 2000-2004, 1 in
19 of the people affected by a climate shock was living in a developing country, compared with
only 1 in 1,500 for OECD countries (UNDP, 2007, p.76). Additionally, the progressively frequency
and intensity of such events might even compromise the historical resilience of some regions
(Lacambra et al., 2008). In this context, understanding the links between natural shocks and human
development —human capital, in particular, becomes important, especially when designing policies
aimed at reducing vulnerability and enhancing the inherent resilience of regions and communities.
To sum up, climate change will increase the risk of exposure to climate shocks, mainly for the
people living in poor countries, and therefore, will become an obstacle to the development goals of
developing countries.
Colombia is a natural disaster hotspot. According to de la Fuente (2012), Colombia ranks at
number 11 in the global ranking of population in areas of risk; 21.2% of its territory is at risk from
two or more natural hazards; 84.7% of the population living in these areas could be potentially
affected; plus 86.6% of the GDP of risk-prone areas is at stake. According to the National System
for the Prevention and Attention of Disasters (SNPAD, 2013), since 1998, there has been an
increasing rate of occurrence of climate shocks events in the country, as well as a raise in the
number of people affected. From 1998 to 2011, the average annual increase in the number of
climate-related events was 23%. Floods and landslides accounted for 50.6% and 20.5% of all
events, respectively. While in 1998 258,341 people were victims of climate-related disasters, in
2010 this figure was 3,319,686.
What is more, the severe weather shocks of 2010 were the worst experienced by Colombia in
its recent history. To mention just a few statistics, with respect to the previous year (2009), the
number of people affected in 2010 increased in 661% (3,319,686), the number of families affected
damaged in 519% (376,349), the number of roads destroyed in 358% (1,104), and the number of
schools affected in 351% (501). The shock was not only intense, but was also felt in almost the
whole territory. In 2009, 513 municipalities were affected by climate events; in 2010 this figure
rose to 1,020 (more than 90% of all Colombian municipalities). Moreover, the shocks persisted in
2011, although to a lesser intensity. The Graph 1 shows the number of people (per 100,000
inhabitants) and the number of municipalities that were affected by climate-related events during
the period 1998-2012. The year 2010 stands alone as the most intense year in terms of the severity
of climate shocks. Therefore, this particularity provides a unique opportunity to conduct a natural
experiment and apply an impact evaluation methodology, such as difference-in-difference, to assess
the impact of this severe change in the intensity of climate events on the schooling achievement of
high school students, as measured by the results in Saber 11 test.
Graph 1. Intensity of Climate Shocks in Colombia, 1998-2012 (number of municipalities
affected times number of people affected by climate-related disasters)
Source: SNPAD, author calculations.
This paper is structured as follows. Section I presents a literature review on the relationship
between climate shocks and human development, it also includes some illustrative empirical studies
relating climate events and human capital, and the importance of perceived future risk in the human
capital investment decisions of households; Section II identifies the main determinants of schooling
outcomes, especially those related to student characteristics, parents’ and peers’ characteristics, and
school and teacher characteristics; Section III introduces the datasets used in this paper and some
summary statistics; Section IV explains the empirical strategy of difference-in-difference estimation
with repeated cross sections; Section V presents the main results; Section VI introduces two
0 200 400 600 800 1000 1200
0 500000 1000000 1500000 2000000 2500000 3000000 3500000
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
possible channels of transmission: health deterioration and schools buildings destruction, and
discusses the possible implications of credit constraints; finally, the last section concludes.
I. Literature Review
A. Climate Shocks and Human Development: Theory
The effects of natural disasters on economic growth (one of the components of human
development) are not clear-cut, since both positive and negative effects are found in the literature,
without providing definite conclusions (Chhibberet al., 2008; Baez et al., 2009; Ferreira and
Schady, 2009; McDermott, 2012). In general, natural disasters reduce the stock of capital in the
economy causing an immediate decrease in the GDP. But, what is the long-term impact of such
disasters on the economic performance of the affected region? On the one hand, authors such as
Cavallo et al. (2010) argue that natural disasters, either large or small, do not seem to have an
apparent impact on the short/long-term economic growth. On the other hand, authors such as
Chhibberet al. (2008), consider the possibility of such impact, by theoretically analyzing four
different yet-to-test scenarios. In the first two scenarios, the long-term growth rate is not affected,
meaning that after the natural shock, the economy will eventually return to its long-term growth
path, either with (scenario 1) or without (scenario 2) a short-term expansion of its production levels.
In the next two scenarios, the long-term growth rate is affected, either negatively (scenario 3),
because of the permanent reduction in the stock of capital, or positively (scenario 4), due to the
technological change introduced by the restitution of capital.
Therefore, the reduction in the stock of capital that results from a natural disaster is likely to
produce a temporary decline in the income and production levels of the affected economy. Now,
what are the possible effects of this chain of events on human capital, especially on schooling
outcomes?
The answer to this question, according to Ferreira and Schady (2009), will depend on the
magnitude of two opposite forces: the income and substitution effects. The income effect, by
reducing households’ available resources, has a negative impact on schooling; while the substitution
effect, by affecting the opportunity cost of studying versus working (with more children studying
after a shock, due to a reduction in child wage), has a positive impact on schooling. As a result, the
total impact of a natural shock on schooling is not clear cut, especially if households are
credit-constrained; however, in the case of poorer countries, the authors claim, the income effect is
(contrary to the case of richer countries). For middle-income countries, like those of Latin America,
empirical evidence suggests that education outcomes are counter-cyclical to economic downturns,
meaning that more children are enrolled in school during economic crisis. Nevertheless, the authors
state that the effects are heterogeneous within and across countries. In this sense, natural shocks
have differential effects depending on gender (women usually suffer the most, Goh, 2012), race,
socioeconomic status, occupation, and location, but the poor are always the most negative affected
(Ferreira and Schady, 2009; Baez et al., 2009).
In fact, climate-related events increase the odds that a household remains or becomes poor
(Glave et al. 2008, for the case of Peru); in fact, these events increase the chances of poverty
persistence (poverty lock-in) and downward mobility (downward consumption trajectories),
hindering the capacity of households for rising to a higher socioeconomic position (Premand and
Vakis, 2010). To this effect, natural disasters (especially floods and droughts) have negative
impacts on both human development (deterioration in the human development index) and poverty
(food poverty, capacities poverty, and asset poverty) (Rodríguez-Oreggia et al., 2010). Moreover,
the long-term effects of these events on human development are felt stronger on poorer regions,
because, even though these regions are more prone to natural catastrophes, they are also less likely
to mobilize reconstruction funds, by, for example, implementing counter-cyclical fiscal policies
(Cavallo and Noy, 2010); plus, these regions usually have lower levels of infrastructural
development, awareness, and coping capacities (Goteng et al., 2012). Accordingly, it is stated that
economic and human development can counteract the negative effects of climate shocks on a
certain region; that is, it increases its resilience (Toya and Skidmore, 2007).
The literature has also acknowledged the existence of direct and indirect effects on human
capital derived from climate-related events. Direct effects include the destruction and depletion of
physical and human capital. One of the immediate consequences of climate shocks is the
destruction of physical capital, such as schools, health centers, households’ assets, and public and
private infrastructure; as well as of human capital, in terms of the casualties, disabilities, illness, and
injuries of students, teachers, and health professionals (Fuentes and Seck, 2007; Baez et al., 2009;
Crespo-Cuaresma, 2010; McDermott, 2011). Wounds and illness keep children from attending
school; death translates into a loss in previous investments in human capital; and disease or
epidemics eruption, which results from contamination or scarcity of water and food supplies,
combined with the favorable conditions for microorganisms to emerge and spread, could
Together, the destruction of physical and human capital increases the marginal cost of
acquiring human capital (Baez et al., 2009), which will negatively affect its future accumulation
and, therefore, the human development possibilities of the affected region.
The negative impacts of the direct effects are certain, but the indirect effects can either
counteract or reinforce these impacts. Contrary to the direct effects, the indirect effects will be
affected by the decisions taken by households after the natural disaster (McDermott, 2012). The loss
of households’ assets, the illness or death of households’ members, which could potentially cut their
available time to generate income, together with the migration and evacuation decisions, will most
probably reduce the family income (Baez et al., 2009; Crespo-Cuaresma, 2010; McDermott, 2011).
Plus, the destruction of infrastructure will require investment decisions by the affected households;
but, poorer families will find it difficult to invest, because of credit restrictions or unavailability of
credit to them. In such situation, credit-constrained households will be forced to disinvest, by
selling-off productive assets, in order to cope with shock. Therefore, this situation will trigger a
vicious circle, since the reduction of productive assets will diminish their ability to generate income
in the future, and this will translate into more vulnerability to future climate shocks (McDermott,
2011). In consequence, when households are credit-constrained, this shock on income will lead
family units to reduce their investment on human and physical capital accumulation. In particular,
the consumption of food, health and educational services will decline. Plus, parents might resort to
children’s time as a buffer mechanism to soften the shock. In this scenario, adding the possible
health impacts derived from the disaster and the possibility that income losses might increase the
opportunity cost of studying, children will be permanent or temporary withdrawn from school (Baez
et al., 2009; McDermott, 2011).
Prices and wages, the amount of parental time, and the discontinuation of schooling are other
indirect channels through which natural disasters affect human capital. The impact of a natural
disaster on prices and wages is unclear, because it will depend on the direction and size of the
income and substitution effects (Baez et al., 2009; Ferreira and Schady, 2009). Additionally, there is
uncertainty about the amount of parental time with children available after a shock, as well as of its
effects on the production of human capital (possibly increasing its marginal cost, Baez et al., 2009).
Finally, because of the discontinuation of schooling, children might not be able to keep up at a later
time or will drop out of the educational system for good, creating a path-dependent effect (Baez et
al., 2009). So, the short-term trade-offs faced by households in order to smooth consumption can
have long-lasting negative effects on the accumulation of human capital, even more when human
(Fuentes and Seck, 2007). In this sense, the evidence supports the fact that the net effect of the
direct and indirect effects is strongly negative and long-lasting (Baez et al., 2009).
B. Climate Shocks and Human Capital: Empirical Evidence
As was stated in the previous section, the occurrence of natural disasters not only increases
mortality risk, affecting the stock of human capital (because of the casualties of educated persons),
but also affects the decisions of families regarding the use of child labor, in order to compensate for
the income losses from the disaster. These hypotheses have been confirmed by different empirical
studies. For example, using Bayesian Model Averaging methods, Crespo-Cuaresma (2010) finds a
strong negative long-run effect of natural disasters on the rate of secondary school enrollment, a
proxy for human capital accumulation. His results are consistent across countries and robust to
different model specifications. The use of child labor to compensate for income losses is evidenced
by Duryea et al. (2007) for the case of Brazil; in this study, the authors analyze the impact of
household economic shocks on the reallocation of children’s time from school to work; their results
indicate that transitory unexpected economic shocks force families to increase children’s labor in
detriment of children’s schooling.
Early life natural disasters have also been documented to have a negative effect on children’s
future human capital. Studying the case of rural Vietnam, Thai and Falaris (2011) find that negative
climate shocks during gestation and early life affect households’ income, by destroying crop
production, and thus, have an indirect effect on children’s nutrition, measured as height-for-age, and
on schooling, measured as delayed entry to school and slower progress once enrolled. The effects
vary by region, which shows dependence on specific region’s constraints, with the regions where
households face greater difficulties to smooth consumption being the most affected by the shock. In
the case of Mozambique, a similar result in terms of children’s nutrition was found by Prado
(2009): Natural disasters negatively affect children’s height-for-age for children between one and
three years old. In the case of Mali, according to De Vreyer et al. (2011),early life shocks, like the
one experienced by Malians in their early childhood back in the period 1987-1989, when a locust
plague hit the country, have a long-lasting effect on nutrition and, thus, on educational enrollment
and completion. The study shows a differential effect on girls and boys, confirming the
gender-discrimination situation of the country, and a lack of insurance mechanisms that could have helped
smoothing consumption.
Another case study, this time for Colombia, reaffirms the negative effects of climate shocks
coffee-growing region of Colombia had a negative short-term impact on schooling and nutrition (the
impact persisted in the medium-term, although with a lesser intensity), with parents reducing their
investments in the human capital of their children. Their findings support the hypothesis that one of
the households’ mechanisms to cope with natural disasters is reducing their investment in their
children’s human capital, by withdrawing them from school or impoverishing their nutrition. Plus,
even when remedial actions are taken right after the shock, a natural disaster might have persistent
effects on the accumulation of human capital, with negative long-term welfare effects.
So far, most of the studies have focused on the negative effects of climate shocks on human
capital. Still, climate shocks can also have positive consequences through a sudden increase in
income. For example, for the case of Indonesian adults, Maccini and Yang (2009) studied the
impact of early-life shocks (higher rainfall) on economic development variables, including health,
education, and household’s assets. Their results point out that higher early-life rainfall has a positive
effect on women’s variables (resulting in higher socioeconomic status), but not on men’s, stressing
the gender discrimination issues of the Indonesian society. The channel through which higher
rainfall influence the future socioeconomic status of women is through its impact on agricultural
production, which, in turn, will increase household’s income, improving, later in life, their health
status and schooling attainment.
C. Perceived Future Climate Risk and Human Capital
Human capital is not only affected by actual climate shocks; the perceived future risk,
whether it materializes or not, can also have a profound impact on this variable. For instance,
evidence from rural Indonesia suggests that parents’ schooling decisions are affected by
environmental risks; more precisely, under riskier environments parents tend to postpone their
children’s entry into school (Korkeala, 2012). Moreover, the effect of a perceived future risky
environment, decomposed into household and village effects, on the stock of human capital of
Indonesian rural children, indicate that village-level risk (the aggregate component), such as past
fluctuations in rainfall, has a negative effect on children’s educational attainment, while
household-level risk (the idiosyncratic component), such as risk in parental income, has not significant effect
(Fitzsimons, 2007). These findings point out the difficulty of households to insure against
village-level risks, and therefore, their need to resort to child labor, in order to buffer against the shock;
these results suggest also that the market for this insurance type might be incomplete (Fitzsimons,
So, focusing only on current shocks and the ex-post responses of households to them might
not provide a full picture of the total costs derived from income risk, as a result of a climate shock,
especially if education exhibits non-linear returns (Kazianga, 2012) or if the effect of future disaster
risk on asset holding is not linear (Yamauchi et al., 2009). On the one hand, the results of Kazianga
(2012) from rural Burkina Faso, although suffering from external validity, indicate that income
uncertainty (income standard deviation) has a negative effect on household schooling decisions.
This is because, in such uncertain environments, parents might decide not to enroll some of their
children in school and put them to work, in order to minimize the impact of future shocks, even if
these shocks do not materialize. On the other hand, if disaster probability is higher than a certain
threshold, then asset holding will be positive; however, future risk has two opposite effects: it
incentives investments, so as to lessen the impacts of future disasters; but it also disincentives
investments, due to the uncertainty of the returns of these investments in front of a future risk
(Yamauchi et al., 2009). In this line of reasoning, the study of Yamauchi et al. (2009), for
Bangladesh, Ethiopia and Malawi, finds that the former effect is greater than the latter in disaster
prone regions. In a similar study, these authors conclude that human capital investments and asset
holding prior to a natural disaster shock help both increasing resilience and upholding investments
in human capital in the aftermath of a shock (Yamauchi et al., 2009b).
II. Determinants of Schooling Achievement
According to the literature, schooling achievement is the result of student characteristics,
parents’ and peers’ characteristics, and school characteristics.
A. Student Characteristics
This section describes the main factors associated to the student personal characteristics that
have been found to be important in determining schooling outcomes, such as test scores. These
characteristics include students’ personality traits, health, time allocation, labor decisions, and
gender.
Students’ personality traits, although not easily measurable, have an important role in
explaining schooling outcomes. For example, Baird (2012) argue that, while for some countries
school characteristics (measured as school resources) still account for a great part of the
performance gap between low and high socioeconomic status students, factors such as students’
effort, interest, and motivation (usually, unmeasured characteristics) are generally more relevant to
self-esteem tend to create virtuous circles of achievement, because, by trusting in their own abilities,
these students put more effort when performing a task, and, therefore, tend to accomplish more,
reinforcing the circle (Darolia and Wydick, 2011). A prior positive academic self-concept (students
who think of themselves as being more able, effective, or confident) have also an important positive
impact on different schooling outcomes, such as interest in the subjects, grades, and scores in
standardized tests (Marsh et al., 2005).
Children’s health has also strong and significant positive causal impact on their academic
outcomes (Wolfe, 1985; Behrman, 1996; Glewwe and Miguel, 2008). One example is provided by
Sabia (2007), who, after controlling for unobservable heterogeneity, finds that obesity, measured
with a body mass index, has a negative impact on schooling outcomes (GPA) of white girls between
14 and 17 years old (the results are less significant for boys and nonwhite girls). Another case is
given by Belot and James (2011), who exploit a shifting towards healthier meal options in
Greenwich’s schools in the UK; their results suggest that a better nutrition, which translates into
better health, improves student’s schooling outcomes and decreases absenteeism.
Similarly, Time allocation and labor decisions have an important effect on schooling
achievement. Focusing on the effects of time allocation of undergraduate students on their academic
results, although the results suffer from endogeneity, Grave (2011) finds that work group and
attending tutorials has a negative impact on grades for below-average students, as well as for
Engineering and Science students; attending courses has a positive effect only for certain groups of
students (high-ability students and females) or certain programs (Engineering and Social Sciences);
while, the time allocated to self-study or working as a tutor or academic assistant is positively
related to higher grades for all students. Now, concerning labor decisions, Montmarquette et al.
(2007) find that working less than 15 hours a week has not necessarily a negative impact on
schooling outcomes; however, students who actually have an intensive work and non-worker
students who show a preference for intensive work, which indicates a predisposition to a paid-job
over studying, are related to low academic achievement (Staff et al., 2010). The preferences of
studying over working have been found to be related to being female, having educated parents, and
attending a private school (Montmarquette et al., 2007).
Finally, gender seems to have an important effect on academic results, which might stem
from inherently gendered behaviors. Niederle and Vesterlund (2010), for example, argue that
differences in the way men and women respond to competitive test-taking settings are responsible
for the observed gender-related gap in mathematics achievement. These difference responses to
women, which should actually be regarded, in the author’s words, as “math skills under competitive
pressure”. In consequence, single-sex education can have a positive impact on females’ math
scores, by influencing their math studying decisions and improving their self-confidence in the
subject (Fryer and Levitt, 2009, cited by Niederle and Vesterlund, 2010).
B. Parents’ and Peers’ Characteristics
People whom students interact with in their daily lives, basically parents and friends, play
also a major role in school achievement.
In the case of rural students, Broomhall and Johnson (1994) analyze the value that people
from rural areas place on education. Their results suggest that the value parents put on schooling
together with the availability of local economic opportunities (or the willingness to find better
opportunities somewhere else) exert an influence on the importance rural students place on
education, which in turn translates into a better student’s performance at school. Therefore, parents’
perception of the benefits their children can reap from education combined with the socioeconomic
conditions of the environment will have a say, in terms of incentives, in the student’s and parent’s
decisions on how much to invest in education, and therefore, on the accumulation of human capital
of individuals.
Apart from family income, which affects student’s academic outcomes, mostly through the
impacts of school choice proxies (Hoxby, 2001, cited by Krieg and Storer, 2008), the literature has
found other parents-related variables that explain to a certain extent schooling performance.
Parent’s cognitive skills, for instance, are highly correlated with their children’s academic
achievement at school; plus, children of parents with higher cognitive skills are likely to perform
better in tests and to have less behavioral and emotional complications (de Coulon et al., 2011).
Literature for college performance indicates that parents’ reward schemes will exert an influence on
their children’s academic effort and on their post-school achievement (Darolia and Wydick, 2011).
Allowances seem to have a negative to neutral effect on these variables, except if the allowance was
given upon the completion of a certain task (conditional allowances), in which case the effect was
positive; purchasing a car in high school relates to a lower academic effort; but, children of parents
who usually recognize their achievements, by influencing their self-esteem, will tend to exert
greater effort in their undergraduate studies (Darolia and Wydick, 2011).
Research on parental closure found mixed effects of this variable on student’s high school
and Todd, 2009). Paternal job loss has a negative impact on children’s schooling outcomes,
although not maternal job loss, which has a non-significant positive effect on the student’s GPA, at
least for the Norwegian case; this difference is explained by the claim that men experience more
mental distress from a job loss experience than women (Rege et al., 2011). Birth order has no
significant effect on test performance, but the number of siblings does negatively impact verbal IQ
results (Steelman and Doby, 1983); the reasons provided by the authors are related to the
importance for language development of parents’ interaction with their children, in terms of
stimulation and attention, which could be considerably affected by the number of siblings that
would have to compete for their parent’s time and dedication.
This latter result was also confirmed by Zimmerman (2003), when analyzing the effects of
peer roommates on test scores. The author’s results suggest that peer roommates have a more
positive and significant effect on verbal tests scores than on math tests scores. Other studies have
also remarked the importance of peers in schooling performance. Focusing on 11 years old British
children, Robertson and Symons (2003) found a strong effect of peer groups, parents’ education,
and social class on children’s academic achievement. Peer effects explain also, to a great extent, the
reasons behind the performance gap between primary school students from Mexico and Cuba (with
Cubans outperforming Mexicans) (McEwan and Marshall, 2004); in this study the socioeconomic
variables of the student’s family appear to be also important, although to a lesser degree; while
school and teacher characteristics have no explanatory power when explaining differences in
academic achievement across nations. Analyzing the United States case, Lee (2007) finds that both
peer racial composition and school have significant effects on the student’s academic outcomes.
C. School Characteristics
Most of the interaction between students and their peers and teachers occur in school
environments, hence the importance of this variable in schooling performance. As in the case of the
number of siblings, the number of students in a class might limit the time a teacher could spend
with each of her pupils. In this respect, some research has found that class size reductions and
teacher density (number of teachers per student) might exert a positive influence on cognitive skills
and academic achievement (Fredriksson and Öckert, 2008; Ding and Lehrer, 2011), but the
evidence is not clear-cut in the case of non-cognitive skills, such as motivation, listening, and
self-concept, in which case family background seems to be more important (Ding and Lehrer, 2011).
Results from a natural experiment and a field experiment suggest that attending a high-scoring
school has a positive impact on the student’s own academic achievement (Hastings and Weinstein,
great part from the socioeconomic characteristics of their students rather than from the school
quality (Krieg and Storer, 2008). The literature points also to the importance of remedial summer
schools and retention programs on academic achievement, especially for young disadvantaged
students (Jacob and Lefgren, 2004).
Teacher quality has been considered an important factor affecting students’ cognitive skills.
In this respect, focusing on public elementary education, Rivkin et al. (2005) argue that teachers’
quality have an important effect on student’s mathematics and reading performance (and therefore
on the school quality), and that being exposed to a higher quality teaching environment can
counterbalance the negative effects of having a low socioeconomic status. However, the authors
found that teacher quality is mostly explained by the unobservable characteristics of the teacher,
since years of experience or having a master degree does not seem to have any significant impact.
At the international level, Glewwe and Kremer (2006) highlight the importance of teacher’s quality
as the most important factor affecting school quality and its cross-country differences. Their claim
is supported by random experiments conducted in different developing countries, where the
substitution of technologies, such as radio education in Nicaragua or computer-assisted learning
programs in India, for weak teachers have had a positive effect on the student’s academic outcomes;
this assertion is also supported by the results found after the implementation of teacher incentives in
countries such as Israel and Kenya, where student’s performance and test scores were significantly
improved (although only on short-run outcomes, in the case of Kenya, and mainly for weaker
students, in the case of Israel).
Lastly, central exit examinations have a positive impact on student’s academic achievement,
(Jürges et al., 2005), but this impact seems to have differential effects, depending on the ability of
students; that is, central exit examinations have a lesser impact on low-ability students, compared to
high-ability students, due to the characteristics of the labor market for less skilled workers (lower
job mobility and limited grades-reading capacity of local employers) (Wößmann, 2005).
III. Data
This paper uses two different unique databases: ICFES database for Saber 11 test and
SNPAD database for natural disasters. The ICFES database contains the test results from the
examination Saber 11, a standardized national test applied to high school Colombian students prior
to graduation. The test is developed by the Instituto Colombiano para el Fomento de la Educación
Superior —ICFES (Colombian Institute for the Promotion of Higher Education). The purpose of the
a year, according to the academic year of the school; however, for most of the institutions the
academic year starts in late January or early February and ends in mid or late November; this
calendar is known as “calendar A”. The test results are required by some universities as part of their
application process; they also serve as a quality indicator that allows comparing the country’s high
schools performance. Now, regarding the contents of the Saber 11 test, the test has two components:
a common core, which evaluates the students’ knowledge in eight (8) different subjects: language
(Spanish), mathematics, biology, chemistry, physics, philosophy, social science, and foreign
language (English); and a flexible core, which allows students to choose one subject out of the six
available options, divided in four in-depth subjects: language, mathematics, biology, or social
science, and two interdisciplinary subjects: environment or violence and society.
This paper uses the Saber 11 (calendar A) database for the period 2008-2012. Specifically, it
uses the following information from the database: test results, student characteristics, and household
characteristics. Test results include the test scores for each of the subjects of the common core
(language, mathematics, biology, chemistry, physics, philosophy, social science, and English) —the
total score was calculated as the arithmetic mean of the common core components. Student
characteristics include: date of birth (three age variables were constructed from this information:
age —the student age when the test was taken, age 15-16 —a dummy variable if the student was 15
or 16 years old, and age-squared); mother education and father education, which ranks the student
parents’ level of education on an ascending scale from 0 to 10 (where 0 means “none education”
and 10 “graduate studies”); sex (male or female), from which the dummy variable male was
created; and work, a dummy variable if the student had a job (paid or not) (the original database had
different job classifications, which were recoded into just one category).
Household characteristics include: social stratum (the social stratification by law, ranging
from 1 to 6 —1 indicating the lowest and 6 the highest, with each strata sharing similar
socioeconomic characteristics; a few students classified as a strata 8 —not stratified, were omitted
from the database); Sisben —El Sistema de Identificación de Potenciales Beneficiarios de
Programas Sociales (The System for the Selection of Beneficiaries of Social Programs), ranging
from level 1 to 4 (level 1, 2, and 3 means that the household is classified in any of these levels,
whereas level 4 includes households that are classified at a different level and those that are not
classified at all), monthly household income (income), ranging from 1 to 7, according to the number
of minimum wages earned by the household unit on a monthly basis (1: less than 1 minimum wage;
2: between 1 and 2; 3: between 2 and 3; 4: between 3 and 5; 5: between 5 and 7; 6: between 7 and
the variable No. of people per dormitory and overcrowding (d) (dummy if the variable No. of people
per dormitory was equal or greater than 2.5) were created; living zone (urban or rural), from which
the dummy variable urban was constructed; and, finally, the dummy variables: car, computer, and
DVD player, each indicating whether or not the household had at least one of these objects, and
Internetconnection and cableTV, each indicating if the household had the service installed (some
variables in the original database specified the number of objects the household had, these were
recoded to “having at least one”).
From the ICFES database, the variable total score is used as dependent variable, whereas the
student and household characteristics are used as controls. The total number of students in the
database for the period 2008-2012 is 2,669,540. Some summary statistics for some of the main
variables as well as the total number of students per year are presented in Table 1.
Table 1. Summary Statistics for ICFES Database Variables, 2008-2012
Category Variable Obs. Mean Std. Dev. Min. Max.
Saber 11 Scores
Total score 2,652,365 44.137 6.447 0.000 87.125
Biology 2,653,386 45.292 8.172 0.000 100.000
Social Science 2,653,386 44.759 9.237 0.000 112.930
Philosophy 2,653,386 40.552 9.265 0.000 84.000
Physics 2,653,386 43.875 8.407 0.000 112.000
English 2,652,365 42.947 10.084 0.000 111.940
Language 2,653,386 45.798 8.320 0.000 93.000
Mathematics 2,653,386 44.763 10.644 0.000 126.000
Chemistry 2,653,386 45.090 7.419 0.000 94.640
Year
2008 508,253 ‐ ‐ ‐ ‐
2009 521,738 ‐ ‐ ‐ ‐
2010 540,452 ‐ ‐ ‐ ‐
2011 540,441 ‐ ‐ ‐ ‐
2012 549,832 ‐ ‐ ‐ ‐
Source: ICFES, Saber 11 database, author calculations.
The ICFES database variables were merged with some variables from the SNPAD national
disasters database. This database was developed by the governmental institution “Sistema Nacional
para la Prevención y Atención de Desastres” (National System for the Prevention and Attention of
Disasters). The database contains the records of the different natural events that have affected
Colombia since 1998 at a municipality level. Some of the variables included in the database are:
date of the event; municipality code; type of event; number of casualties; number of people
affected, wounded, or missing; number of houses destroyed or damaged; and number of different
public infrastructure affected.
In order to create a shock variable (shock), the following variables were used from the
construction of the variable shock is detailed as follows. For each year and for each Colombian
municipality, the number of people affected by natural disasters per 100,000 inhabitants was created
(see Map 1 for the year 2010). This variable was used afterwards to calculate the average of the
annual gross changes between 2004 and 2009, as well as the gross change between 2009 and 2010.
The 2004-2009 average was then compared with the 2009-2010 gross change using the percentiles
50 and 99. This information was used to select the treatment and control groups. A municipality
was considered treated with intensity 1 if the 2009-2010 gross change was greater than the
percentile 50 but less than the percentile 99 of the 2004-2009 average (shock=1), and it was
considered treated with intensity 2 if the 2009-2010 gross change was greater than the percentile 99
(shock=2) of the 2004-2009 average. If the 2009-2010 gross change was less than or equal to the
percentile 50 of the 2004-2009 average, the municipality was included in the control group.
The reason why the shock variables were used as indicators of treatment is because the
impact of the natural disasters on the test scores does not seem to follow a linear pattern (as
suggested by Sachs (2006) in terms of the non-linear effects of climate change). In fact, the impact
of the variable “number of people affected by 100,000 inhabitants”, when included as a explanatory
variable of the Saber 11 test scores, was close to zero and non significant, indicating that a natural
shock might affect cognitive skills only if it surpasses a certain threshold.
Map 1. People Affected by Climate-Related Events in Colombia, 2010 (Per 100,000
Source: SNPAD, author calculations.
Table 2 shows some summary statistics for the SNPAD database. In particular, it shows the
number of people affected by climate-related disasters, per 100,000 inhabitants, as well as the total
number of people affected. The source of the municipalities’ populations between 2006 and 2012
was “Departamento Administrativo Nacional de Estadística” DANE (National Administrative
Department of Statistics).
Table 2. Summary Statistics for SNPAD Database Variables, 2006-2012
Variable Mean Std. Dev. Min. Max.
No. of people affected Disasters 2006 3,100.3 9,149.0 0 135,656.8 711,447
Disasters 2007 6,271.9 22,320.0 0 275,423.9 1,559,377
Disasters 2008 8,932.6 22,490 0 214,432.4 1,877,504
Disasters 2009 2,274.7 12,593.3 0 345,651.6 435,851
Disasters 2010 16,614.01 31,605.4 0 577,616.1 3,319,686
Disasters 2011 11,263.16 26,875.1 0 348,439.5 2,178,557
Disasters 2012 1,560.78 7,421.7 0 128,402.6 282,333 Source: SNPAD, DANE, author calculations.
IV. Empirical Strategy: Difference-in-Difference Estimation with Repeated Cross Sections
This paper uses a difference-in-difference estimation with repeated cross sections to measure
the impact of the climate shocks of 2010 and 2011 on the test scores of the Saber 11 test in the
period 2010-2012. The dummy variable shock indicates the treatment status of the individuals, that
is, a student will belong to the treatment group if she lives in a municipality for which the change in
the number of people affected by climate-related disasters (per 100,000 inhabitants) between 2009
and 2010 was either greater than the percentile 50 (shock intensity 1) or greater than the percentile
99 (shock intensity 2) of the average of the annual changes between 2004 and 2009. The student
will belong to the control group if she lives in a municipality where the change in the number of
people affected by climate-related disasters (per 100,000 inhabitants) between 2009 and 2010 was
less than or equal to the median (percentile 50) of the average of the annual changes between 2004
and 2009 (shock=0). Even though shock varies at the municipality level, this paper uses student i as
the unit of observation. The reason for this is the possibility to control for observables available in
the ICFES database, as well as to examine heterogeneous effects. The baseline model is given by
equation (1), whereas the time-heterogeneous effects model is given by equation (2):
2011 ∗ 1
2010 2011 2012 ∗ 2010
∗ 2011 ∗ 2012 2008 2
where the outcome variable represents the test score of student i in time t; is a
dummy variable equal to 1 if the student lives in a municipality j where the shock intensity was 1,
and equal to 2 if the shock intensity was 2 ; post is a dummy variable equal to one in the post-shock
period (2010-2012); y2010, y2011, and y2012 represent dummy variables for the years 2010, 2011,
and 2012; is a vector of student and parents control variables (age, age 15-16 —dummy for ages
15 or 16, age-squared, mother education, father education, male —dummy for student’s sex, and
work —dummy equal to one if the student works), household control variables (social stratum,
sisben, income —monthly household income, No. of people per dormitory, overcrowding (d) —
dummy equal to one if No. of people per dormitory is equal or greater than 2.5, urban —dummy
equal to one if the student lives in an urban area, car —dummy equal to one if the household has at
least one car, computer —dummy equal to one if the household has at least one computer, DVD
dummy equal to one if the household has Internet connection, and cable TV, —dummy equal to one
if the household has cable TV); is a vector of climate shocks variables for the years pre and post
2010 (disasters 2006, disasters 2007, disasters 2008, disasters 2009, disasters 2011, disasters
2012), representing the number of people affected by climate disaster per 100,000 inhabitants;
is a vector of dummy variables to control for 2008, 2010, and 2011 year effects; y2008 is a
dummy variable for the year 2008; represents school fixed effects; and, finally, is an error
term which satisfies | 2010 0.
The difference-in-difference estimator ( ) in equation (1) will be given by ; whereas the
differential effect of years 2010, 2011, and 2012 in equation (2) will be given by ( ),
( ), and ( ). Under the baseline model specification, β1 represents the treatment
group specific effect; β2 is a time trend, which is common to treatment and control groups; and β3 is
the true effect of treatment. In order for the model to be correctly estimated, the following
assumptions are required: (1) the error term must have mean zero, (2) the error term must not be
correlated with any of the variables in the equation, and (3) the parallel-trend assumption, which
guarantees that in the absence of treatment (shock), the average change in the test score for the
treatment group would have been the same as the average change for the control group. Trends in
the average Saber 11 scores for the years before the shock (2008, and 2009) for both treatment and
control groups are presented in Graph 2. Before the shock, the trends in the average scores were
similar for both groups; however, after the shock, the treatment group exhibited a different path.
Graphically, the strongest effects of the 2010-2011 natural events were felt in 2011; but they were
Graph 2. Parallel-Trend Assumption
Source: ICFES, SNPAD, author calculations.
This paper implements also a triple-difference approach, in order to analyze the
heterogeneous effects, as the literature has pointed out that the impacts of climate shocks on
schooling outcomes vary according to certain characteristics, such as sex or living area, for
example. The triple-difference model specification for the baseline model is given by equation (3),
whereas the triple-difference for the time-heterogeneous effects model is given by equation (4); in
both model specifications, represents a variable for which the expected outcome varies with:
∗ ∗ ∗
∗ ∗ 3
2010 2011 2012 ∗
∗ 2010 ∗ 2011 ∗ 2012 ∗ 2010
∗ 2011 ∗ 2012 ∗ ∗ 2010 ∗
∗ 2011 ∗ ∗ 2012 2008 4
This paper estimates equations (3) and (4) to measure the differential impacts of shocks of
2010-2011 (shock) on the variables male, urban, disasters 2006, disasters 2007, disasters 2008, disasters
2009, social stratum, and income. The triple-differences estimator ( ) in equation (6) is given
by ; measures the impact of the variable shock when Z = 0, and the impact when Z
= 1; and so the estimator identifies the differential impact of the variable shock for Z =
40 41 42 43 44 45 46 47
2008 2009 2010 2011 2012
1 with respect to Z = 0. In the case of equation (4), the differential impact of shock on the different
values of Z for the years 2010, 2011, and 2012 is given by ( ), ( ), and
( ), respectively.
IV. Results
Graph 3 presents the difference-in-difference estimations of , , and ,
using two different model specifications of equation (2): (1) difference-in-difference without
controls and school fixed effects and (2) difference-in-difference with controls and school fixed
effects. In general, the estimators are quite similar in both cases. However, the negative effect gets a
little bit stronger as controls and fixed effects are added to the simple difference-in-difference
estimation. This fact would imply that the estimation of for the different post-years, without
such controls and fixed effects, might be biased; that is, the control variables and the fixed effects
help explain both the test scores and the fact that the municipality had suffered more from the
2010-2011 climate-related shocks than the national municipality average. However, the estimator would
be biased downwards. For example, studying in a school with poor infrastructure increases the
chances of being affected by a landslide or a flood, augmenting the possibility of being treated, but
it also relates to a lower test score. By the same token, a high household income decreases the
chances of being affected, perhaps by living in a house better equipped to annual floods or by
having access to credit and insurance markets, but it also has a positive effect on the test scores.
Therefore, the effect of the omitted variables on the treatment indicator and on the outcome variable
seems to follow opposite directions, and so not including control variables would underestimate the
impact estimator.
Graph 3. Impact of the Climate Shocks of 2010 on the Saber 11 Test Scores Using (1)
Source: ICFES, SNPAD, author calculations.
Note: All estimators are statistically significant at 1%. Standard errors were clustered at the school level.
The results shows that the climate shocks of 2010, as measured by the variable shock, had an
important and significant impact on students’ Saber 11 tests scores; the shock was more strongly
felt in 2011, but its intensity had repercussions even in 2012, although to a lesser extent. Despite the
fact that the climate shocks were intense in 2010, these weather-events did not have such great
impact on the 2010 test results, compared to 2011 and 2012; the reason for this is that the 2010
Saber 11 test was taken on September 12th 2010, but (1) most of the natural disasters concentrated
in just two months: August and November, and (2) in terms of the breadth and depth of the
disasters, November stood alone as the time of the year when most municipalities suffered the most
from these events.
Table 3 presents the complete results of the estimation of equations (1) and (2) using pooled
OLS with clustered standard errors at the school level. Students’ and parents’ characteristics, as
expected, have an important impact on the test score, result confirmed by the literature review of
Section II, as well as by the studies on the determinants of academic performance in Colombia
(Chica et al., 2011; Gaviria and Barrientos, 2001). Students’ age is negatively related to test scores,
but having the correct age for grade has positive impact on this outcome; having a job while
studying is related to a lower test performance, and so is being female. All of these variables are
important in the estimation and might help capture some of the students’ unobserved skills and
traits, such as interest or motivation, as well as their possible time allocation. Parents’ level of
education is also important, as it is related to higher levels of income; and by having higher salaries
or belonging to a higher socioeconomic class, parents can have access to higher quality education, ‐0.204***
‐0.708***
‐0.436*** ‐0.26***
‐0.722***
‐0.45***
‐0.8 ‐0.7 ‐0.6 ‐0.5 ‐0.4 ‐0.3 ‐0.2 ‐0.1 0
2010 2011 2012
to a wider range of school choice (Gaviria and Barrientos, 2001; Krieg and Storer, 2008), and to
school complements, such as personal teachers or assistive technology.
Now, concerning student’s household characteristics, living in an overcrowded house (more
than 2.5 people sharing a bedroom) has a slightly negative effect on test scores, possibly because
the student might have more siblings or relatives living in the same place and so she might not
receive enough parents’ dedication and attention. Living in an urban area is positively correlated
with a better scoring, since urban dwellers might not only enjoy a broader range of academic
services, but put a greater value on education too, because the benefits of investing in education, in
terms of the economic opportunities available after graduation, are broader if the student lives in an
urban area. In this sense, rural students might put a lower value on education, as a result of the
shortage of high quality schools, and the lack of proper incentives to have a good score, as argued
by Broomhall and Johnson (1994). Having at least one computer and Internet connection has a
positive impact on schooling outcomes; these household services can act as a complement of the
education received at school, but also as a substitute for weak teaching (Glewwe and Kremer,
2006). In contrast, having at least one car, one DVD player, or cable TV is related to a negative
score; cable TV and DVD player can take up time that would have otherwise been allocated to
studying, while the negative effect of having at least one car could be related to the parents reward
schemes discussed in Section II (Darolia and Wydick, 2011).
Table 3. Impact of the Climate Shocks of 2010 on the Saber 11 Test Scores, 2010- 2012
Dependent variable: Score Pooled OLS
Baseline Model [eq.
(1)]
Time-Heterogeneous
Effects Model [eq. (2)]
Shock.1 0.095
(0.051)
0.098 (0.051)
Shock.2 0.297**
(0.096)
0.302** (0.096)
Shock.1*Post -0.296***
(0.038)
Shock.2*Post -1.017***
(0.044)
Post 0.925***
(0.032)
Shock.1*y2010 -0.180***
(0.035)
Shock.2*y2010 -0.554***
(0.042)
Shock.1*y2011 -0.408***
(0.05)
Shock.2*y2011 -1.548***
(0.058)
Shock.1*yt2012 -0.306***
(0.044)
Shock.2*y2012 -0.955***
(0.051)
(0.008) (0.008)
Age 15-16 1.280***
(0.01)
1.280*** (0.01)
Age-squared 0.002***
(0.000)
0.002*** (0.000)
Mother education 0.186***
(0.003)
0.186*** (0.003)
Father education 0.191***
(0.002)
0.191*** (0.002)
Social stratum 0.150***
(0.009)
0.150*** (0.009)
Overcrowding 0.017**
(0.006)
0.017** (0.006)
Overcrowding -0.069***
(0.012)
-0.068*** (0.012)
Income 0.272***
(0.006)
0.271*** (0.006)
Work -0.236***
(0.015)
-0.236*** (0.015)
Male 1.084***
(0.01)
1.084*** (0.01)
Sisben 0.136***
(0.005)
0.136*** (0.005)
Urban 0.490***
(0.016)
0.490*** (0.016)
Car -0.395***
(0.012)
-0.395*** (0.012)
Computer 0.372***
(0.011)
0.373*** (0.011)
DVD player -0.251***
(0.009)
-0.249*** (0.009)
Internet connection 0.114***
(0.012)
0.110*** (0.012)
Cable TV -0.248***
(0.01)
-0.249*** (0.01)
y2008 -0.194***
(0.018)
-0.195*** (0.018)
y2010 -0.116***
(0.015)
0.668*** (0.029)
y2011 -0.403***
(0.016)
0.678*** (0.039)
y2012 0.920***
(0.036)
Disasters 2006 0.000
(0.000)
0.000 (0.000)
Disasters 2007 0.000
(0.000)
0.000 (0.000)
Disasters 2008 -0.000*
(0.000)
-0.000* (0.000)
Disasters 2009 0.000
(0.000)
0.000 (0.000)
Disasters 2010 0.000
(0.000)
0.000 (0.000)
Disasters 2011 0.000
(0.000)
0.000 (0.000)
Disasters 2012 -0.000*
(0.000)
-0.000* (0.000)
Constant 41.551***
(0.118)
41.544*** (0.118)
School Fixed Effects YES YES
R-squared 0.371 0.371
Obs. 2,422,673 2,422,673
Source: ICFES, SNPAD, author calculations.
Notes: (1) *p<0.05, **p<0.01, ***p<0.001, (2) standard errors in parenthesis are clustered at the school level.
Equations (1) and (2) were also estimated having as dependent variable each of the core
subjects of the test. Table 4 presents the difference-in-difference estimator of the impact of the 2010
climate shocks on each of the core subjects evaluated in the Saber 11 test. In general, the effect of
the shocks was negative and highly significant (except for Chemistry and English). As expected, the
impact was greater in municipalities that experienced a stronger shock (shock=2). The most affected
subjects were Philosophy, Language, Mathematics, and Biology; Language and Philosophy
followed a similar trajectory, which could be an indicator that these subjects require similar skills
and that these skills could have been in turn heavily affected by the climate shocks of 2010.
According to ICFES, Language, Philosophy, and Social Science tests evaluate students’ skills in
interpretation, argumentation, and proposition; the reason why the effect was stronger in the first
two subjects than in Social Science could be related to the fact that Philosophy might be closer to
Language, since it is aimed at improving students’ critical thinking, communication, and reading
abilities. The literature review suggested that the lack of interaction between students’ and their
parents, peers, and teachers have a robust impact on the development of their language skills; then,
one possible explanation for the strong effect of the climate disasters on Language and Philosophy
is that these shocks could have prevented such interactions.
Relatively to other subjects, the effect of the shocks on Mathematics was important (it was
the most affected subject in 2010). The skills evaluated in this subject are: communication,
reasoning, and problem-solving. In the case of Natural Sciences (Biology, Chemistry, and Physics),
which evaluate the identification, enquiring, and explanation skills, the strength of the effects was
different for each subject, being stronger for Biology, and weaker for Chemistry, which was the
second least affected of all common core subjects after English. The shock effect on Social Science,
which assesses the same skills as Language and Philosophy, was comparatively low; while the
impact on English, which evaluates grammar skills, textual skills, and textual coherence, was
generally not significant and had a positive sign in 2011.
As was stated by Niederle and Vesterlund (2010) in Section II, males tend to outperform
females in mathematics. This tendency is also confirmed in this study; in fact, although males
outperform females in all subjects, except Philosophy, Mathematics exhibits the greatest score
differences between males and females, followed by Physics, Social Science, and Biology. Girls
tend to perform better in Philosophy, whereas the advantage of boys over girls in Language is the
lowest among the boys-dominated subjects. This might provide new evidence that performance in